2 * QDM2 compatible decoder
3 * Copyright (c) 2003 Ewald Snel
4 * Copyright (c) 2005 Benjamin Larsson
5 * Copyright (c) 2005 Alex Beregszaszi
6 * Copyright (c) 2005 Roberto Togni
8 * This file is part of FFmpeg.
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 * @author Ewald Snel, Benjamin Larsson, Alex Beregszaszi, Roberto Togni
30 * The decoder is not perfect yet, there are still some distortions
31 * especially on files encoded with 16 or 8 subbands.
38 #include "libavutil/channel_layout.h"
39 #include "libavutil/mem_internal.h"
40 #include "libavutil/thread.h"
42 #define BITSTREAM_READER_LE
45 #include "bytestream.h"
47 #include "mpegaudio.h"
48 #include "mpegaudiodsp.h"
51 #include "qdm2_tablegen.h"
53 #define QDM2_LIST_ADD(list, size, packet) \
56 list[size - 1].next = &list[size]; \
58 list[size].packet = packet; \
59 list[size].next = NULL; \
63 // Result is 8, 16 or 30
64 #define QDM2_SB_USED(sub_sampling) (((sub_sampling) >= 2) ? 30 : 8 << (sub_sampling))
66 #define FIX_NOISE_IDX(noise_idx) \
67 if ((noise_idx) >= 3840) \
68 (noise_idx) -= 3840; \
70 #define SB_DITHERING_NOISE(sb,noise_idx) (noise_table[(noise_idx)++] * sb_noise_attenuation[(sb)])
72 #define SAMPLES_NEEDED \
73 av_log (NULL,AV_LOG_INFO,"This file triggers some untested code. Please contact the developers.\n");
75 #define SAMPLES_NEEDED_2(why) \
76 av_log (NULL,AV_LOG_INFO,"This file triggers some missing code. Please contact the developers.\nPosition: %s\n",why);
78 #define QDM2_MAX_FRAME_SIZE 512
80 typedef int8_t sb_int8_array[2][30][64];
85 typedef struct QDM2SubPacket {
86 int type; ///< subpacket type
87 unsigned int size; ///< subpacket size
88 const uint8_t *data; ///< pointer to subpacket data (points to input data buffer, it's not a private copy)
92 * A node in the subpacket list
94 typedef struct QDM2SubPNode {
95 QDM2SubPacket *packet; ///< packet
96 struct QDM2SubPNode *next; ///< pointer to next packet in the list, NULL if leaf node
99 typedef struct QDM2Complex {
104 typedef struct FFTTone {
106 QDM2Complex *complex;
115 typedef struct FFTCoefficient {
123 typedef struct QDM2FFT {
124 DECLARE_ALIGNED(32, QDM2Complex, complex)[MPA_MAX_CHANNELS][256];
128 * QDM2 decoder context
130 typedef struct QDM2Context {
131 /// Parameters from codec header, do not change during playback
132 int nb_channels; ///< number of channels
133 int channels; ///< number of channels
134 int group_size; ///< size of frame group (16 frames per group)
135 int fft_size; ///< size of FFT, in complex numbers
136 int checksum_size; ///< size of data block, used also for checksum
138 /// Parameters built from header parameters, do not change during playback
139 int group_order; ///< order of frame group
140 int fft_order; ///< order of FFT (actually fftorder+1)
141 int frame_size; ///< size of data frame
143 int sub_sampling; ///< subsampling: 0=25%, 1=50%, 2=100% */
144 int coeff_per_sb_select; ///< selector for "num. of coeffs. per subband" tables. Can be 0, 1, 2
145 int cm_table_select; ///< selector for "coding method" tables. Can be 0, 1 (from init: 0-4)
147 /// Packets and packet lists
148 QDM2SubPacket sub_packets[16]; ///< the packets themselves
149 QDM2SubPNode sub_packet_list_A[16]; ///< list of all packets
150 QDM2SubPNode sub_packet_list_B[16]; ///< FFT packets B are on list
151 int sub_packets_B; ///< number of packets on 'B' list
152 QDM2SubPNode sub_packet_list_C[16]; ///< packets with errors?
153 QDM2SubPNode sub_packet_list_D[16]; ///< DCT packets
156 FFTTone fft_tones[1000];
159 FFTCoefficient fft_coefs[1000];
161 int fft_coefs_min_index[5];
162 int fft_coefs_max_index[5];
163 int fft_level_exp[6];
164 RDFTContext rdft_ctx;
168 const uint8_t *compressed_data;
170 float output_buffer[QDM2_MAX_FRAME_SIZE * MPA_MAX_CHANNELS * 2];
173 MPADSPContext mpadsp;
174 DECLARE_ALIGNED(32, float, synth_buf)[MPA_MAX_CHANNELS][512*2];
175 int synth_buf_offset[MPA_MAX_CHANNELS];
176 DECLARE_ALIGNED(32, float, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT];
177 DECLARE_ALIGNED(32, float, samples)[MPA_MAX_CHANNELS * MPA_FRAME_SIZE];
179 /// Mixed temporary data used in decoding
180 float tone_level[MPA_MAX_CHANNELS][30][64];
181 int8_t coding_method[MPA_MAX_CHANNELS][30][64];
182 int8_t quantized_coeffs[MPA_MAX_CHANNELS][10][8];
183 int8_t tone_level_idx_base[MPA_MAX_CHANNELS][30][8];
184 int8_t tone_level_idx_hi1[MPA_MAX_CHANNELS][3][8][8];
185 int8_t tone_level_idx_mid[MPA_MAX_CHANNELS][26][8];
186 int8_t tone_level_idx_hi2[MPA_MAX_CHANNELS][26];
187 int8_t tone_level_idx[MPA_MAX_CHANNELS][30][64];
188 int8_t tone_level_idx_temp[MPA_MAX_CHANNELS][30][64];
191 int has_errors; ///< packet has errors
192 int superblocktype_2_3; ///< select fft tables and some algorithm based on superblock type
193 int do_synth_filter; ///< used to perform or skip synthesis filter
196 int noise_idx; ///< index for dithering noise table
199 static const int switchtable[23] = {
200 0, 5, 1, 5, 5, 5, 5, 5, 2, 5, 5, 5, 5, 5, 5, 5, 3, 5, 5, 5, 5, 5, 4
203 static int qdm2_get_vlc(GetBitContext *gb, const VLC *vlc, int flag, int depth)
207 value = get_vlc2(gb, vlc->table, vlc->bits, depth);
209 /* stage-2, 3 bits exponent escape sequence */
211 value = get_bits(gb, get_bits(gb, 3) + 1);
213 /* stage-3, optional */
218 av_log(NULL, AV_LOG_ERROR, "value %d in qdm2_get_vlc too large\n", value);
222 tmp= vlc_stage3_values[value];
224 if ((value & ~3) > 0)
225 tmp += get_bits(gb, (value >> 2));
232 static int qdm2_get_se_vlc(const VLC *vlc, GetBitContext *gb, int depth)
234 int value = qdm2_get_vlc(gb, vlc, 0, depth);
236 return (value & 1) ? ((value + 1) >> 1) : -(value >> 1);
242 * @param data pointer to data to be checksummed
243 * @param length data length
244 * @param value checksum value
246 * @return 0 if checksum is OK
248 static uint16_t qdm2_packet_checksum(const uint8_t *data, int length, int value)
252 for (i = 0; i < length; i++)
255 return (uint16_t)(value & 0xffff);
259 * Fill a QDM2SubPacket structure with packet type, size, and data pointer.
261 * @param gb bitreader context
262 * @param sub_packet packet under analysis
264 static void qdm2_decode_sub_packet_header(GetBitContext *gb,
265 QDM2SubPacket *sub_packet)
267 sub_packet->type = get_bits(gb, 8);
269 if (sub_packet->type == 0) {
270 sub_packet->size = 0;
271 sub_packet->data = NULL;
273 sub_packet->size = get_bits(gb, 8);
275 if (sub_packet->type & 0x80) {
276 sub_packet->size <<= 8;
277 sub_packet->size |= get_bits(gb, 8);
278 sub_packet->type &= 0x7f;
281 if (sub_packet->type == 0x7f)
282 sub_packet->type |= (get_bits(gb, 8) << 8);
284 // FIXME: this depends on bitreader-internal data
285 sub_packet->data = &gb->buffer[get_bits_count(gb) / 8];
288 av_log(NULL, AV_LOG_DEBUG, "Subpacket: type=%d size=%d start_offs=%x\n",
289 sub_packet->type, sub_packet->size, get_bits_count(gb) / 8);
293 * Return node pointer to first packet of requested type in list.
295 * @param list list of subpackets to be scanned
296 * @param type type of searched subpacket
297 * @return node pointer for subpacket if found, else NULL
299 static QDM2SubPNode *qdm2_search_subpacket_type_in_list(QDM2SubPNode *list,
302 while (list && list->packet) {
303 if (list->packet->type == type)
311 * Replace 8 elements with their average value.
312 * Called by qdm2_decode_superblock before starting subblock decoding.
316 static void average_quantized_coeffs(QDM2Context *q)
318 int i, j, n, ch, sum;
320 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
322 for (ch = 0; ch < q->nb_channels; ch++)
323 for (i = 0; i < n; i++) {
326 for (j = 0; j < 8; j++)
327 sum += q->quantized_coeffs[ch][i][j];
333 for (j = 0; j < 8; j++)
334 q->quantized_coeffs[ch][i][j] = sum;
339 * Build subband samples with noise weighted by q->tone_level.
340 * Called by synthfilt_build_sb_samples.
343 * @param sb subband index
345 static void build_sb_samples_from_noise(QDM2Context *q, int sb)
349 FIX_NOISE_IDX(q->noise_idx);
354 for (ch = 0; ch < q->nb_channels; ch++) {
355 for (j = 0; j < 64; j++) {
356 q->sb_samples[ch][j * 2][sb] =
357 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
358 q->sb_samples[ch][j * 2 + 1][sb] =
359 SB_DITHERING_NOISE(sb, q->noise_idx) * q->tone_level[ch][sb][j];
365 * Called while processing data from subpackets 11 and 12.
366 * Used after making changes to coding_method array.
368 * @param sb subband index
369 * @param channels number of channels
370 * @param coding_method q->coding_method[0][0][0]
372 static int fix_coding_method_array(int sb, int channels,
373 sb_int8_array coding_method)
379 for (ch = 0; ch < channels; ch++) {
380 for (j = 0; j < 64; ) {
381 if (coding_method[ch][sb][j] < 8)
383 if ((coding_method[ch][sb][j] - 8) > 22) {
387 switch (switchtable[coding_method[ch][sb][j] - 8]) {
411 for (k = 0; k < run; k++) {
413 int sbjk = sb + (j + k) / 64;
418 if (coding_method[ch][sbjk][(j + k) % 64] > coding_method[ch][sb][j]) {
421 //not debugged, almost never used
422 memset(&coding_method[ch][sb][j + k], case_val,
424 memset(&coding_method[ch][sb][j + k], case_val,
437 * Related to synthesis filter
438 * Called by process_subpacket_10
441 * @param flag 1 if called after getting data from subpacket 10, 0 if no subpacket 10
443 static void fill_tone_level_array(QDM2Context *q, int flag)
445 int i, sb, ch, sb_used;
448 for (ch = 0; ch < q->nb_channels; ch++)
449 for (sb = 0; sb < 30; sb++)
450 for (i = 0; i < 8; i++) {
451 if ((tab=coeff_per_sb_for_dequant[q->coeff_per_sb_select][sb]) < (last_coeff[q->coeff_per_sb_select] - 1))
452 tmp = q->quantized_coeffs[ch][tab + 1][i] * dequant_table[q->coeff_per_sb_select][tab + 1][sb]+
453 q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
455 tmp = q->quantized_coeffs[ch][tab][i] * dequant_table[q->coeff_per_sb_select][tab][sb];
458 q->tone_level_idx_base[ch][sb][i] = (tmp / 256) & 0xff;
461 sb_used = QDM2_SB_USED(q->sub_sampling);
463 if ((q->superblocktype_2_3 != 0) && !flag) {
464 for (sb = 0; sb < sb_used; sb++)
465 for (ch = 0; ch < q->nb_channels; ch++)
466 for (i = 0; i < 64; i++) {
467 q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
468 if (q->tone_level_idx[ch][sb][i] < 0)
469 q->tone_level[ch][sb][i] = 0;
471 q->tone_level[ch][sb][i] = fft_tone_level_table[0][q->tone_level_idx[ch][sb][i] & 0x3f];
474 tab = q->superblocktype_2_3 ? 0 : 1;
475 for (sb = 0; sb < sb_used; sb++) {
476 if ((sb >= 4) && (sb <= 23)) {
477 for (ch = 0; ch < q->nb_channels; ch++)
478 for (i = 0; i < 64; i++) {
479 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
480 q->tone_level_idx_hi1[ch][sb / 8][i / 8][i % 8] -
481 q->tone_level_idx_mid[ch][sb - 4][i / 8] -
482 q->tone_level_idx_hi2[ch][sb - 4];
483 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
484 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
485 q->tone_level[ch][sb][i] = 0;
487 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
491 for (ch = 0; ch < q->nb_channels; ch++)
492 for (i = 0; i < 64; i++) {
493 tmp = q->tone_level_idx_base[ch][sb][i / 8] -
494 q->tone_level_idx_hi1[ch][2][i / 8][i % 8] -
495 q->tone_level_idx_hi2[ch][sb - 4];
496 q->tone_level_idx[ch][sb][i] = tmp & 0xff;
497 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
498 q->tone_level[ch][sb][i] = 0;
500 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
503 for (ch = 0; ch < q->nb_channels; ch++)
504 for (i = 0; i < 64; i++) {
505 tmp = q->tone_level_idx[ch][sb][i] = q->tone_level_idx_base[ch][sb][i / 8];
506 if ((tmp < 0) || (!q->superblocktype_2_3 && !tmp))
507 q->tone_level[ch][sb][i] = 0;
509 q->tone_level[ch][sb][i] = fft_tone_level_table[tab][tmp & 0x3f];
518 * Related to synthesis filter
519 * Called by process_subpacket_11
520 * c is built with data from subpacket 11
521 * Most of this function is used only if superblock_type_2_3 == 0,
522 * never seen it in samples.
524 * @param tone_level_idx
525 * @param tone_level_idx_temp
526 * @param coding_method q->coding_method[0][0][0]
527 * @param nb_channels number of channels
528 * @param c coming from subpacket 11, passed as 8*c
529 * @param superblocktype_2_3 flag based on superblock packet type
530 * @param cm_table_select q->cm_table_select
532 static void fill_coding_method_array(sb_int8_array tone_level_idx,
533 sb_int8_array tone_level_idx_temp,
534 sb_int8_array coding_method,
536 int c, int superblocktype_2_3,
540 int tmp, acc, esp_40, comp;
541 int add1, add2, add3, add4;
544 if (!superblocktype_2_3) {
545 /* This case is untested, no samples available */
546 avpriv_request_sample(NULL, "!superblocktype_2_3");
548 for (ch = 0; ch < nb_channels; ch++) {
549 for (sb = 0; sb < 30; sb++) {
550 for (j = 1; j < 63; j++) { // The loop only iterates to 63 so the code doesn't overflow the buffer
551 add1 = tone_level_idx[ch][sb][j] - 10;
554 add2 = add3 = add4 = 0;
556 add2 = tone_level_idx[ch][sb - 2][j] + tone_level_idx_offset_table[sb][0] - 6;
561 add3 = tone_level_idx[ch][sb - 1][j] + tone_level_idx_offset_table[sb][1] - 6;
566 add4 = tone_level_idx[ch][sb + 1][j] + tone_level_idx_offset_table[sb][3] - 6;
570 tmp = tone_level_idx[ch][sb][j + 1] * 2 - add4 - add3 - add2 - add1;
573 tone_level_idx_temp[ch][sb][j + 1] = tmp & 0xff;
575 tone_level_idx_temp[ch][sb][0] = tone_level_idx_temp[ch][sb][1];
579 for (ch = 0; ch < nb_channels; ch++)
580 for (sb = 0; sb < 30; sb++)
581 for (j = 0; j < 64; j++)
582 acc += tone_level_idx_temp[ch][sb][j];
584 multres = 0x66666667LL * (acc * 10);
585 esp_40 = (multres >> 32) / 8 + ((multres & 0xffffffff) >> 31);
586 for (ch = 0; ch < nb_channels; ch++)
587 for (sb = 0; sb < 30; sb++)
588 for (j = 0; j < 64; j++) {
589 comp = tone_level_idx_temp[ch][sb][j]* esp_40 * 10;
592 comp /= 256; // signed shift
620 coding_method[ch][sb][j] = ((tmp & 0xfffa) + 30 )& 0xff;
622 for (sb = 0; sb < 30; sb++)
623 fix_coding_method_array(sb, nb_channels, coding_method);
624 for (ch = 0; ch < nb_channels; ch++)
625 for (sb = 0; sb < 30; sb++)
626 for (j = 0; j < 64; j++)
628 if (coding_method[ch][sb][j] < 10)
629 coding_method[ch][sb][j] = 10;
632 if (coding_method[ch][sb][j] < 16)
633 coding_method[ch][sb][j] = 16;
635 if (coding_method[ch][sb][j] < 30)
636 coding_method[ch][sb][j] = 30;
639 } else { // superblocktype_2_3 != 0
640 for (ch = 0; ch < nb_channels; ch++)
641 for (sb = 0; sb < 30; sb++)
642 for (j = 0; j < 64; j++)
643 coding_method[ch][sb][j] = coding_method_table[cm_table_select][sb];
648 * Called by process_subpacket_11 to process more data from subpacket 11
650 * Called by process_subpacket_12 to process data from subpacket 12 with
654 * @param gb bitreader context
655 * @param length packet length in bits
656 * @param sb_min lower subband processed (sb_min included)
657 * @param sb_max higher subband processed (sb_max excluded)
659 static int synthfilt_build_sb_samples(QDM2Context *q, GetBitContext *gb,
660 int length, int sb_min, int sb_max)
662 int sb, j, k, n, ch, run, channels;
663 int joined_stereo, zero_encoding;
665 float type34_div = 0;
666 float type34_predictor;
668 int sign_bits[16] = {0};
671 // If no data use noise
672 for (sb=sb_min; sb < sb_max; sb++)
673 build_sb_samples_from_noise(q, sb);
678 for (sb = sb_min; sb < sb_max; sb++) {
679 channels = q->nb_channels;
681 if (q->nb_channels <= 1 || sb < 12)
686 joined_stereo = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
689 if (get_bits_left(gb) >= 16)
690 for (j = 0; j < 16; j++)
691 sign_bits[j] = get_bits1(gb);
693 for (j = 0; j < 64; j++)
694 if (q->coding_method[1][sb][j] > q->coding_method[0][sb][j])
695 q->coding_method[0][sb][j] = q->coding_method[1][sb][j];
697 if (fix_coding_method_array(sb, q->nb_channels,
699 av_log(NULL, AV_LOG_ERROR, "coding method invalid\n");
700 build_sb_samples_from_noise(q, sb);
706 for (ch = 0; ch < channels; ch++) {
707 FIX_NOISE_IDX(q->noise_idx);
708 zero_encoding = (get_bits_left(gb) >= 1) ? get_bits1(gb) : 0;
709 type34_predictor = 0.0;
712 for (j = 0; j < 128; ) {
713 switch (q->coding_method[ch][sb][j / 2]) {
715 if (get_bits_left(gb) >= 10) {
717 for (k = 0; k < 5; k++) {
718 if ((j + 2 * k) >= 128)
720 samples[2 * k] = get_bits1(gb) ? dequant_1bit[joined_stereo][2 * get_bits1(gb)] : 0;
725 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
726 return AVERROR_INVALIDDATA;
729 for (k = 0; k < 5; k++)
730 samples[2 * k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
732 for (k = 0; k < 5; k++)
733 samples[2 * k + 1] = SB_DITHERING_NOISE(sb,q->noise_idx);
735 for (k = 0; k < 10; k++)
736 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
742 if (get_bits_left(gb) >= 1) {
747 f -= noise_samples[((sb + 1) * (j +5 * ch + 1)) & 127] * 9.0 / 40.0;
750 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
756 if (get_bits_left(gb) >= 10) {
758 for (k = 0; k < 5; k++) {
761 samples[k] = (get_bits1(gb) == 0) ? 0 : dequant_1bit[joined_stereo][2 * get_bits1(gb)];
764 n = get_bits (gb, 8);
766 av_log(NULL, AV_LOG_ERROR, "Invalid 8bit codeword\n");
767 return AVERROR_INVALIDDATA;
770 for (k = 0; k < 5; k++)
771 samples[k] = dequant_1bit[joined_stereo][random_dequant_index[n][k]];
774 for (k = 0; k < 5; k++)
775 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
781 if (get_bits_left(gb) >= 7) {
784 av_log(NULL, AV_LOG_ERROR, "Invalid 7bit codeword\n");
785 return AVERROR_INVALIDDATA;
788 for (k = 0; k < 3; k++)
789 samples[k] = (random_dequant_type24[n][k] - 2.0) * 0.5;
791 for (k = 0; k < 3; k++)
792 samples[k] = SB_DITHERING_NOISE(sb,q->noise_idx);
798 if (get_bits_left(gb) >= 4) {
799 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type30, 0, 1);
800 if (index >= FF_ARRAY_ELEMS(type30_dequant)) {
801 av_log(NULL, AV_LOG_ERROR, "index %d out of type30_dequant array\n", index);
802 return AVERROR_INVALIDDATA;
804 samples[0] = type30_dequant[index];
806 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
812 if (get_bits_left(gb) >= 7) {
814 type34_div = (float)(1 << get_bits(gb, 2));
815 samples[0] = ((float)get_bits(gb, 5) - 16.0) / 15.0;
816 type34_predictor = samples[0];
819 unsigned index = qdm2_get_vlc(gb, &vlc_tab_type34, 0, 1);
820 if (index >= FF_ARRAY_ELEMS(type34_delta)) {
821 av_log(NULL, AV_LOG_ERROR, "index %d out of type34_delta array\n", index);
822 return AVERROR_INVALIDDATA;
824 samples[0] = type34_delta[index] / type34_div + type34_predictor;
825 type34_predictor = samples[0];
828 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
834 samples[0] = SB_DITHERING_NOISE(sb,q->noise_idx);
840 for (k = 0; k < run && j + k < 128; k++) {
841 q->sb_samples[0][j + k][sb] =
842 q->tone_level[0][sb][(j + k) / 2] * samples[k];
843 if (q->nb_channels == 2) {
844 if (sign_bits[(j + k) / 8])
845 q->sb_samples[1][j + k][sb] =
846 q->tone_level[1][sb][(j + k) / 2] * -samples[k];
848 q->sb_samples[1][j + k][sb] =
849 q->tone_level[1][sb][(j + k) / 2] * samples[k];
853 for (k = 0; k < run; k++)
855 q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k];
866 * Init the first element of a channel in quantized_coeffs with data
867 * from packet 10 (quantized_coeffs[ch][0]).
868 * This is similar to process_subpacket_9, but for a single channel
869 * and for element [0]
870 * same VLC tables as process_subpacket_9 are used.
872 * @param quantized_coeffs pointer to quantized_coeffs[ch][0]
873 * @param gb bitreader context
875 static int init_quantized_coeffs_elem0(int8_t *quantized_coeffs,
878 int i, k, run, level, diff;
880 if (get_bits_left(gb) < 16)
882 level = qdm2_get_vlc(gb, &vlc_tab_level, 0, 2);
884 quantized_coeffs[0] = level;
886 for (i = 0; i < 7; ) {
887 if (get_bits_left(gb) < 16)
889 run = qdm2_get_vlc(gb, &vlc_tab_run, 0, 1) + 1;
894 if (get_bits_left(gb) < 16)
896 diff = qdm2_get_se_vlc(&vlc_tab_diff, gb, 2);
898 for (k = 1; k <= run; k++)
899 quantized_coeffs[i + k] = (level + ((k * diff) / run));
908 * Related to synthesis filter, process data from packet 10
909 * Init part of quantized_coeffs via function init_quantized_coeffs_elem0
910 * Init tone_level_idx_hi1, tone_level_idx_hi2, tone_level_idx_mid with
911 * data from packet 10
914 * @param gb bitreader context
916 static void init_tone_level_dequantization(QDM2Context *q, GetBitContext *gb)
920 for (ch = 0; ch < q->nb_channels; ch++) {
921 init_quantized_coeffs_elem0(q->quantized_coeffs[ch][0], gb);
923 if (get_bits_left(gb) < 16) {
924 memset(q->quantized_coeffs[ch][0], 0, 8);
929 n = q->sub_sampling + 1;
931 for (sb = 0; sb < n; sb++)
932 for (ch = 0; ch < q->nb_channels; ch++)
933 for (j = 0; j < 8; j++) {
934 if (get_bits_left(gb) < 1)
937 for (k=0; k < 8; k++) {
938 if (get_bits_left(gb) < 16)
940 q->tone_level_idx_hi1[ch][sb][j][k] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi1, 0, 2);
943 for (k=0; k < 8; k++)
944 q->tone_level_idx_hi1[ch][sb][j][k] = 0;
948 n = QDM2_SB_USED(q->sub_sampling) - 4;
950 for (sb = 0; sb < n; sb++)
951 for (ch = 0; ch < q->nb_channels; ch++) {
952 if (get_bits_left(gb) < 16)
954 q->tone_level_idx_hi2[ch][sb] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_hi2, 0, 2);
956 q->tone_level_idx_hi2[ch][sb] -= 16;
958 for (j = 0; j < 8; j++)
959 q->tone_level_idx_mid[ch][sb][j] = -16;
962 n = QDM2_SB_USED(q->sub_sampling) - 5;
964 for (sb = 0; sb < n; sb++)
965 for (ch = 0; ch < q->nb_channels; ch++)
966 for (j = 0; j < 8; j++) {
967 if (get_bits_left(gb) < 16)
969 q->tone_level_idx_mid[ch][sb][j] = qdm2_get_vlc(gb, &vlc_tab_tone_level_idx_mid, 0, 2) - 32;
974 * Process subpacket 9, init quantized_coeffs with data from it
977 * @param node pointer to node with packet
979 static int process_subpacket_9(QDM2Context *q, QDM2SubPNode *node)
982 int i, j, k, n, ch, run, level, diff;
984 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
986 n = coeff_per_sb_for_avg[q->coeff_per_sb_select][QDM2_SB_USED(q->sub_sampling) - 1] + 1;
988 for (i = 1; i < n; i++)
989 for (ch = 0; ch < q->nb_channels; ch++) {
990 level = qdm2_get_vlc(&gb, &vlc_tab_level, 0, 2);
991 q->quantized_coeffs[ch][i][0] = level;
993 for (j = 0; j < (8 - 1); ) {
994 run = qdm2_get_vlc(&gb, &vlc_tab_run, 0, 1) + 1;
995 diff = qdm2_get_se_vlc(&vlc_tab_diff, &gb, 2);
1000 for (k = 1; k <= run; k++)
1001 q->quantized_coeffs[ch][i][j + k] = (level + ((k * diff) / run));
1008 for (ch = 0; ch < q->nb_channels; ch++)
1009 for (i = 0; i < 8; i++)
1010 q->quantized_coeffs[ch][0][i] = 0;
1016 * Process subpacket 10 if not null, else
1019 * @param node pointer to node with packet
1021 static void process_subpacket_10(QDM2Context *q, QDM2SubPNode *node)
1026 init_get_bits(&gb, node->packet->data, node->packet->size * 8);
1027 init_tone_level_dequantization(q, &gb);
1028 fill_tone_level_array(q, 1);
1030 fill_tone_level_array(q, 0);
1035 * Process subpacket 11
1038 * @param node pointer to node with packet
1040 static void process_subpacket_11(QDM2Context *q, QDM2SubPNode *node)
1046 length = node->packet->size * 8;
1047 init_get_bits(&gb, node->packet->data, length);
1051 int c = get_bits(&gb, 13);
1054 fill_coding_method_array(q->tone_level_idx,
1055 q->tone_level_idx_temp, q->coding_method,
1056 q->nb_channels, 8 * c,
1057 q->superblocktype_2_3, q->cm_table_select);
1060 synthfilt_build_sb_samples(q, &gb, length, 0, 8);
1064 * Process subpacket 12
1067 * @param node pointer to node with packet
1069 static void process_subpacket_12(QDM2Context *q, QDM2SubPNode *node)
1075 length = node->packet->size * 8;
1076 init_get_bits(&gb, node->packet->data, length);
1079 synthfilt_build_sb_samples(q, &gb, length, 8, QDM2_SB_USED(q->sub_sampling));
1083 * Process new subpackets for synthesis filter
1086 * @param list list with synthesis filter packets (list D)
1088 static void process_synthesis_subpackets(QDM2Context *q, QDM2SubPNode *list)
1090 QDM2SubPNode *nodes[4];
1092 nodes[0] = qdm2_search_subpacket_type_in_list(list, 9);
1094 process_subpacket_9(q, nodes[0]);
1096 nodes[1] = qdm2_search_subpacket_type_in_list(list, 10);
1098 process_subpacket_10(q, nodes[1]);
1100 process_subpacket_10(q, NULL);
1102 nodes[2] = qdm2_search_subpacket_type_in_list(list, 11);
1103 if (nodes[0] && nodes[1] && nodes[2])
1104 process_subpacket_11(q, nodes[2]);
1106 process_subpacket_11(q, NULL);
1108 nodes[3] = qdm2_search_subpacket_type_in_list(list, 12);
1109 if (nodes[0] && nodes[1] && nodes[3])
1110 process_subpacket_12(q, nodes[3]);
1112 process_subpacket_12(q, NULL);
1116 * Decode superblock, fill packet lists.
1120 static void qdm2_decode_super_block(QDM2Context *q)
1123 QDM2SubPacket header, *packet;
1124 int i, packet_bytes, sub_packet_size, sub_packets_D;
1125 unsigned int next_index = 0;
1127 memset(q->tone_level_idx_hi1, 0, sizeof(q->tone_level_idx_hi1));
1128 memset(q->tone_level_idx_mid, 0, sizeof(q->tone_level_idx_mid));
1129 memset(q->tone_level_idx_hi2, 0, sizeof(q->tone_level_idx_hi2));
1131 q->sub_packets_B = 0;
1134 average_quantized_coeffs(q); // average elements in quantized_coeffs[max_ch][10][8]
1136 init_get_bits(&gb, q->compressed_data, q->compressed_size * 8);
1137 qdm2_decode_sub_packet_header(&gb, &header);
1139 if (header.type < 2 || header.type >= 8) {
1141 av_log(NULL, AV_LOG_ERROR, "bad superblock type\n");
1145 q->superblocktype_2_3 = (header.type == 2 || header.type == 3);
1146 packet_bytes = (q->compressed_size - get_bits_count(&gb) / 8);
1148 init_get_bits(&gb, header.data, header.size * 8);
1150 if (header.type == 2 || header.type == 4 || header.type == 5) {
1151 int csum = 257 * get_bits(&gb, 8);
1152 csum += 2 * get_bits(&gb, 8);
1154 csum = qdm2_packet_checksum(q->compressed_data, q->checksum_size, csum);
1158 av_log(NULL, AV_LOG_ERROR, "bad packet checksum\n");
1163 q->sub_packet_list_B[0].packet = NULL;
1164 q->sub_packet_list_D[0].packet = NULL;
1166 for (i = 0; i < 6; i++)
1167 if (--q->fft_level_exp[i] < 0)
1168 q->fft_level_exp[i] = 0;
1170 for (i = 0; packet_bytes > 0; i++) {
1173 if (i >= FF_ARRAY_ELEMS(q->sub_packet_list_A)) {
1174 SAMPLES_NEEDED_2("too many packet bytes");
1178 q->sub_packet_list_A[i].next = NULL;
1181 q->sub_packet_list_A[i - 1].next = &q->sub_packet_list_A[i];
1183 /* seek to next block */
1184 init_get_bits(&gb, header.data, header.size * 8);
1185 skip_bits(&gb, next_index * 8);
1187 if (next_index >= header.size)
1191 /* decode subpacket */
1192 packet = &q->sub_packets[i];
1193 qdm2_decode_sub_packet_header(&gb, packet);
1194 next_index = packet->size + get_bits_count(&gb) / 8;
1195 sub_packet_size = ((packet->size > 0xff) ? 1 : 0) + packet->size + 2;
1197 if (packet->type == 0)
1200 if (sub_packet_size > packet_bytes) {
1201 if (packet->type != 10 && packet->type != 11 && packet->type != 12)
1203 packet->size += packet_bytes - sub_packet_size;
1206 packet_bytes -= sub_packet_size;
1208 /* add subpacket to 'all subpackets' list */
1209 q->sub_packet_list_A[i].packet = packet;
1211 /* add subpacket to related list */
1212 if (packet->type == 8) {
1213 SAMPLES_NEEDED_2("packet type 8");
1215 } else if (packet->type >= 9 && packet->type <= 12) {
1216 /* packets for MPEG Audio like Synthesis Filter */
1217 QDM2_LIST_ADD(q->sub_packet_list_D, sub_packets_D, packet);
1218 } else if (packet->type == 13) {
1219 for (j = 0; j < 6; j++)
1220 q->fft_level_exp[j] = get_bits(&gb, 6);
1221 } else if (packet->type == 14) {
1222 for (j = 0; j < 6; j++)
1223 q->fft_level_exp[j] = qdm2_get_vlc(&gb, &fft_level_exp_vlc, 0, 2);
1224 } else if (packet->type == 15) {
1225 SAMPLES_NEEDED_2("packet type 15")
1227 } else if (packet->type >= 16 && packet->type < 48 &&
1228 !fft_subpackets[packet->type - 16]) {
1229 /* packets for FFT */
1230 QDM2_LIST_ADD(q->sub_packet_list_B, q->sub_packets_B, packet);
1232 } // Packet bytes loop
1234 if (q->sub_packet_list_D[0].packet) {
1235 process_synthesis_subpackets(q, q->sub_packet_list_D);
1236 q->do_synth_filter = 1;
1237 } else if (q->do_synth_filter) {
1238 process_subpacket_10(q, NULL);
1239 process_subpacket_11(q, NULL);
1240 process_subpacket_12(q, NULL);
1244 static void qdm2_fft_init_coefficient(QDM2Context *q, int sub_packet,
1245 int offset, int duration, int channel,
1248 if (q->fft_coefs_min_index[duration] < 0)
1249 q->fft_coefs_min_index[duration] = q->fft_coefs_index;
1251 q->fft_coefs[q->fft_coefs_index].sub_packet =
1252 ((sub_packet >= 16) ? (sub_packet - 16) : sub_packet);
1253 q->fft_coefs[q->fft_coefs_index].channel = channel;
1254 q->fft_coefs[q->fft_coefs_index].offset = offset;
1255 q->fft_coefs[q->fft_coefs_index].exp = exp;
1256 q->fft_coefs[q->fft_coefs_index].phase = phase;
1257 q->fft_coefs_index++;
1260 static void qdm2_fft_decode_tones(QDM2Context *q, int duration,
1261 GetBitContext *gb, int b)
1263 int channel, stereo, phase, exp;
1264 int local_int_4, local_int_8, stereo_phase, local_int_10;
1265 int local_int_14, stereo_exp, local_int_20, local_int_28;
1271 local_int_8 = (4 - duration);
1272 local_int_10 = 1 << (q->group_order - duration - 1);
1275 while (get_bits_left(gb)>0) {
1276 if (q->superblocktype_2_3) {
1277 while ((n = qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2)) < 2) {
1278 if (get_bits_left(gb)<0) {
1279 if(local_int_4 < q->group_size)
1280 av_log(NULL, AV_LOG_ERROR, "overread in qdm2_fft_decode_tones()\n");
1285 local_int_4 += local_int_10;
1286 local_int_28 += (1 << local_int_8);
1288 local_int_4 += 8 * local_int_10;
1289 local_int_28 += (8 << local_int_8);
1294 if (local_int_10 <= 2) {
1295 av_log(NULL, AV_LOG_ERROR, "qdm2_fft_decode_tones() stuck\n");
1298 offset += qdm2_get_vlc(gb, &vlc_tab_fft_tone_offset[local_int_8], 1, 2);
1299 while (offset >= (local_int_10 - 1)) {
1300 offset += (1 - (local_int_10 - 1));
1301 local_int_4 += local_int_10;
1302 local_int_28 += (1 << local_int_8);
1306 if (local_int_4 >= q->group_size)
1309 local_int_14 = (offset >> local_int_8);
1310 if (local_int_14 >= FF_ARRAY_ELEMS(fft_level_index_table))
1313 if (q->nb_channels > 1) {
1314 channel = get_bits1(gb);
1315 stereo = get_bits1(gb);
1321 exp = qdm2_get_vlc(gb, (b ? &fft_level_exp_vlc : &fft_level_exp_alt_vlc), 0, 2);
1322 exp += q->fft_level_exp[fft_level_index_table[local_int_14]];
1323 exp = (exp < 0) ? 0 : exp;
1325 phase = get_bits(gb, 3);
1330 stereo_exp = (exp - qdm2_get_vlc(gb, &fft_stereo_exp_vlc, 0, 1));
1331 stereo_phase = (phase - qdm2_get_vlc(gb, &fft_stereo_phase_vlc, 0, 1));
1332 if (stereo_phase < 0)
1336 if (q->frequency_range > (local_int_14 + 1)) {
1337 int sub_packet = (local_int_20 + local_int_28);
1339 if (q->fft_coefs_index + stereo >= FF_ARRAY_ELEMS(q->fft_coefs))
1342 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1343 channel, exp, phase);
1345 qdm2_fft_init_coefficient(q, sub_packet, offset, duration,
1347 stereo_exp, stereo_phase);
1353 static void qdm2_decode_fft_packets(QDM2Context *q)
1355 int i, j, min, max, value, type, unknown_flag;
1358 if (!q->sub_packet_list_B[0].packet)
1361 /* reset minimum indexes for FFT coefficients */
1362 q->fft_coefs_index = 0;
1363 for (i = 0; i < 5; i++)
1364 q->fft_coefs_min_index[i] = -1;
1366 /* process subpackets ordered by type, largest type first */
1367 for (i = 0, max = 256; i < q->sub_packets_B; i++) {
1368 QDM2SubPacket *packet = NULL;
1370 /* find subpacket with largest type less than max */
1371 for (j = 0, min = 0; j < q->sub_packets_B; j++) {
1372 value = q->sub_packet_list_B[j].packet->type;
1373 if (value > min && value < max) {
1375 packet = q->sub_packet_list_B[j].packet;
1381 /* check for errors (?) */
1386 (packet->type < 16 || packet->type >= 48 ||
1387 fft_subpackets[packet->type - 16]))
1390 /* decode FFT tones */
1391 init_get_bits(&gb, packet->data, packet->size * 8);
1393 if (packet->type >= 32 && packet->type < 48 && !fft_subpackets[packet->type - 16])
1398 type = packet->type;
1400 if ((type >= 17 && type < 24) || (type >= 33 && type < 40)) {
1401 int duration = q->sub_sampling + 5 - (type & 15);
1403 if (duration >= 0 && duration < 4)
1404 qdm2_fft_decode_tones(q, duration, &gb, unknown_flag);
1405 } else if (type == 31) {
1406 for (j = 0; j < 4; j++)
1407 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1408 } else if (type == 46) {
1409 for (j = 0; j < 6; j++)
1410 q->fft_level_exp[j] = get_bits(&gb, 6);
1411 for (j = 0; j < 4; j++)
1412 qdm2_fft_decode_tones(q, j, &gb, unknown_flag);
1414 } // Loop on B packets
1416 /* calculate maximum indexes for FFT coefficients */
1417 for (i = 0, j = -1; i < 5; i++)
1418 if (q->fft_coefs_min_index[i] >= 0) {
1420 q->fft_coefs_max_index[j] = q->fft_coefs_min_index[i];
1424 q->fft_coefs_max_index[j] = q->fft_coefs_index;
1427 static void qdm2_fft_generate_tone(QDM2Context *q, FFTTone *tone)
1432 const double iscale = 2.0 * M_PI / 512.0;
1434 tone->phase += tone->phase_shift;
1436 /* calculate current level (maximum amplitude) of tone */
1437 level = fft_tone_envelope_table[tone->duration][tone->time_index] * tone->level;
1438 c.im = level * sin(tone->phase * iscale);
1439 c.re = level * cos(tone->phase * iscale);
1441 /* generate FFT coefficients for tone */
1442 if (tone->duration >= 3 || tone->cutoff >= 3) {
1443 tone->complex[0].im += c.im;
1444 tone->complex[0].re += c.re;
1445 tone->complex[1].im -= c.im;
1446 tone->complex[1].re -= c.re;
1448 f[1] = -tone->table[4];
1449 f[0] = tone->table[3] - tone->table[0];
1450 f[2] = 1.0 - tone->table[2] - tone->table[3];
1451 f[3] = tone->table[1] + tone->table[4] - 1.0;
1452 f[4] = tone->table[0] - tone->table[1];
1453 f[5] = tone->table[2];
1454 for (i = 0; i < 2; i++) {
1455 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].re +=
1457 tone->complex[fft_cutoff_index_table[tone->cutoff][i]].im +=
1458 c.im * ((tone->cutoff <= i) ? -f[i] : f[i]);
1460 for (i = 0; i < 4; i++) {
1461 tone->complex[i].re += c.re * f[i + 2];
1462 tone->complex[i].im += c.im * f[i + 2];
1466 /* copy the tone if it has not yet died out */
1467 if (++tone->time_index < ((1 << (5 - tone->duration)) - 1)) {
1468 memcpy(&q->fft_tones[q->fft_tone_end], tone, sizeof(FFTTone));
1469 q->fft_tone_end = (q->fft_tone_end + 1) % 1000;
1473 static void qdm2_fft_tone_synthesizer(QDM2Context *q, int sub_packet)
1476 const double iscale = 0.25 * M_PI;
1478 for (ch = 0; ch < q->channels; ch++) {
1479 memset(q->fft.complex[ch], 0, q->fft_size * sizeof(QDM2Complex));
1483 /* apply FFT tones with duration 4 (1 FFT period) */
1484 if (q->fft_coefs_min_index[4] >= 0)
1485 for (i = q->fft_coefs_min_index[4]; i < q->fft_coefs_max_index[4]; i++) {
1489 if (q->fft_coefs[i].sub_packet != sub_packet)
1492 ch = (q->channels == 1) ? 0 : q->fft_coefs[i].channel;
1493 level = (q->fft_coefs[i].exp < 0) ? 0.0 : fft_tone_level_table[q->superblocktype_2_3 ? 0 : 1][q->fft_coefs[i].exp & 63];
1495 c.re = level * cos(q->fft_coefs[i].phase * iscale);
1496 c.im = level * sin(q->fft_coefs[i].phase * iscale);
1497 q->fft.complex[ch][q->fft_coefs[i].offset + 0].re += c.re;
1498 q->fft.complex[ch][q->fft_coefs[i].offset + 0].im += c.im;
1499 q->fft.complex[ch][q->fft_coefs[i].offset + 1].re -= c.re;
1500 q->fft.complex[ch][q->fft_coefs[i].offset + 1].im -= c.im;
1503 /* generate existing FFT tones */
1504 for (i = q->fft_tone_end; i != q->fft_tone_start; ) {
1505 qdm2_fft_generate_tone(q, &q->fft_tones[q->fft_tone_start]);
1506 q->fft_tone_start = (q->fft_tone_start + 1) % 1000;
1509 /* create and generate new FFT tones with duration 0 (long) to 3 (short) */
1510 for (i = 0; i < 4; i++)
1511 if (q->fft_coefs_min_index[i] >= 0) {
1512 for (j = q->fft_coefs_min_index[i]; j < q->fft_coefs_max_index[i]; j++) {
1516 if (q->fft_coefs[j].sub_packet != sub_packet)
1520 offset = q->fft_coefs[j].offset >> four_i;
1521 ch = (q->channels == 1) ? 0 : q->fft_coefs[j].channel;
1523 if (offset < q->frequency_range) {
1525 tone.cutoff = offset;
1527 tone.cutoff = (offset >= 60) ? 3 : 2;
1529 tone.level = (q->fft_coefs[j].exp < 0) ? 0.0 : fft_tone_level_table[q->superblocktype_2_3 ? 0 : 1][q->fft_coefs[j].exp & 63];
1530 tone.complex = &q->fft.complex[ch][offset];
1531 tone.table = fft_tone_sample_table[i][q->fft_coefs[j].offset - (offset << four_i)];
1532 tone.phase = 64 * q->fft_coefs[j].phase - (offset << 8) - 128;
1533 tone.phase_shift = (2 * q->fft_coefs[j].offset + 1) << (7 - four_i);
1535 tone.time_index = 0;
1537 qdm2_fft_generate_tone(q, &tone);
1540 q->fft_coefs_min_index[i] = j;
1544 static void qdm2_calculate_fft(QDM2Context *q, int channel, int sub_packet)
1546 const float gain = (q->channels == 1 && q->nb_channels == 2) ? 0.5f : 1.0f;
1547 float *out = q->output_buffer + channel;
1549 q->fft.complex[channel][0].re *= 2.0f;
1550 q->fft.complex[channel][0].im = 0.0f;
1551 q->rdft_ctx.rdft_calc(&q->rdft_ctx, (FFTSample *)q->fft.complex[channel]);
1552 /* add samples to output buffer */
1553 for (i = 0; i < FFALIGN(q->fft_size, 8); i++) {
1554 out[0] += q->fft.complex[channel][i].re * gain;
1555 out[q->channels] += q->fft.complex[channel][i].im * gain;
1556 out += 2 * q->channels;
1562 * @param index subpacket number
1564 static void qdm2_synthesis_filter(QDM2Context *q, int index)
1566 int i, k, ch, sb_used, sub_sampling, dither_state = 0;
1568 /* copy sb_samples */
1569 sb_used = QDM2_SB_USED(q->sub_sampling);
1571 for (ch = 0; ch < q->channels; ch++)
1572 for (i = 0; i < 8; i++)
1573 for (k = sb_used; k < SBLIMIT; k++)
1574 q->sb_samples[ch][(8 * index) + i][k] = 0;
1576 for (ch = 0; ch < q->nb_channels; ch++) {
1577 float *samples_ptr = q->samples + ch;
1579 for (i = 0; i < 8; i++) {
1580 ff_mpa_synth_filter_float(&q->mpadsp,
1581 q->synth_buf[ch], &(q->synth_buf_offset[ch]),
1582 ff_mpa_synth_window_float, &dither_state,
1583 samples_ptr, q->nb_channels,
1584 q->sb_samples[ch][(8 * index) + i]);
1585 samples_ptr += 32 * q->nb_channels;
1589 /* add samples to output buffer */
1590 sub_sampling = (4 >> q->sub_sampling);
1592 for (ch = 0; ch < q->channels; ch++)
1593 for (i = 0; i < q->frame_size; i++)
1594 q->output_buffer[q->channels * i + ch] += (1 << 23) * q->samples[q->nb_channels * sub_sampling * i + ch];
1598 * Init static data (does not depend on specific file)
1600 static av_cold void qdm2_init_static_data(void) {
1602 softclip_table_init();
1604 init_noise_samples();
1606 ff_mpa_synth_init_float();
1610 * Init parameters from codec extradata
1612 static av_cold int qdm2_decode_init(AVCodecContext *avctx)
1614 static AVOnce init_static_once = AV_ONCE_INIT;
1615 QDM2Context *s = avctx->priv_data;
1616 int tmp_val, tmp, size;
1619 /* extradata parsing
1628 32 size (including this field)
1630 32 type (=QDM2 or QDMC)
1632 32 size (including this field, in bytes)
1633 32 tag (=QDCA) // maybe mandatory parameters
1636 32 samplerate (=44100)
1638 32 block size (=4096)
1639 32 frame size (=256) (for one channel)
1640 32 packet size (=1300)
1642 32 size (including this field, in bytes)
1643 32 tag (=QDCP) // maybe some tuneable parameters
1653 if (!avctx->extradata || (avctx->extradata_size < 48)) {
1654 av_log(avctx, AV_LOG_ERROR, "extradata missing or truncated\n");
1655 return AVERROR_INVALIDDATA;
1658 bytestream2_init(&gb, avctx->extradata, avctx->extradata_size);
1660 while (bytestream2_get_bytes_left(&gb) > 8) {
1661 if (bytestream2_peek_be64(&gb) == (((uint64_t)MKBETAG('f','r','m','a') << 32) |
1662 (uint64_t)MKBETAG('Q','D','M','2')))
1664 bytestream2_skip(&gb, 1);
1667 if (bytestream2_get_bytes_left(&gb) < 12) {
1668 av_log(avctx, AV_LOG_ERROR, "not enough extradata (%i)\n",
1669 bytestream2_get_bytes_left(&gb));
1670 return AVERROR_INVALIDDATA;
1673 bytestream2_skip(&gb, 8);
1674 size = bytestream2_get_be32(&gb);
1676 if (size > bytestream2_get_bytes_left(&gb)) {
1677 av_log(avctx, AV_LOG_ERROR, "extradata size too small, %i < %i\n",
1678 bytestream2_get_bytes_left(&gb), size);
1679 return AVERROR_INVALIDDATA;
1682 av_log(avctx, AV_LOG_DEBUG, "size: %d\n", size);
1683 if (bytestream2_get_be32(&gb) != MKBETAG('Q','D','C','A')) {
1684 av_log(avctx, AV_LOG_ERROR, "invalid extradata, expecting QDCA\n");
1685 return AVERROR_INVALIDDATA;
1688 bytestream2_skip(&gb, 4);
1690 avctx->channels = s->nb_channels = s->channels = bytestream2_get_be32(&gb);
1691 if (s->channels <= 0 || s->channels > MPA_MAX_CHANNELS) {
1692 av_log(avctx, AV_LOG_ERROR, "Invalid number of channels\n");
1693 return AVERROR_INVALIDDATA;
1695 avctx->channel_layout = avctx->channels == 2 ? AV_CH_LAYOUT_STEREO :
1698 avctx->sample_rate = bytestream2_get_be32(&gb);
1699 avctx->bit_rate = bytestream2_get_be32(&gb);
1700 s->group_size = bytestream2_get_be32(&gb);
1701 s->fft_size = bytestream2_get_be32(&gb);
1702 s->checksum_size = bytestream2_get_be32(&gb);
1703 if (s->checksum_size >= 1U << 28 || s->checksum_size <= 1) {
1704 av_log(avctx, AV_LOG_ERROR, "data block size invalid (%u)\n", s->checksum_size);
1705 return AVERROR_INVALIDDATA;
1708 s->fft_order = av_log2(s->fft_size) + 1;
1710 // Fail on unknown fft order
1711 if ((s->fft_order < 7) || (s->fft_order > 9)) {
1712 avpriv_request_sample(avctx, "Unknown FFT order %d", s->fft_order);
1713 return AVERROR_PATCHWELCOME;
1716 // something like max decodable tones
1717 s->group_order = av_log2(s->group_size) + 1;
1718 s->frame_size = s->group_size / 16; // 16 iterations per super block
1720 if (s->frame_size > QDM2_MAX_FRAME_SIZE)
1721 return AVERROR_INVALIDDATA;
1723 s->sub_sampling = s->fft_order - 7;
1724 s->frequency_range = 255 / (1 << (2 - s->sub_sampling));
1726 if (s->frame_size * 4 >> s->sub_sampling > MPA_FRAME_SIZE) {
1727 avpriv_request_sample(avctx, "large frames");
1728 return AVERROR_PATCHWELCOME;
1731 switch ((s->sub_sampling * 2 + s->channels - 1)) {
1732 case 0: tmp = 40; break;
1733 case 1: tmp = 48; break;
1734 case 2: tmp = 56; break;
1735 case 3: tmp = 72; break;
1736 case 4: tmp = 80; break;
1737 case 5: tmp = 100;break;
1738 default: tmp=s->sub_sampling; break;
1741 if ((tmp * 1000) < avctx->bit_rate) tmp_val = 1;
1742 if ((tmp * 1440) < avctx->bit_rate) tmp_val = 2;
1743 if ((tmp * 1760) < avctx->bit_rate) tmp_val = 3;
1744 if ((tmp * 2240) < avctx->bit_rate) tmp_val = 4;
1745 s->cm_table_select = tmp_val;
1747 if (avctx->bit_rate <= 8000)
1748 s->coeff_per_sb_select = 0;
1749 else if (avctx->bit_rate < 16000)
1750 s->coeff_per_sb_select = 1;
1752 s->coeff_per_sb_select = 2;
1754 if (s->fft_size != (1 << (s->fft_order - 1))) {
1755 av_log(avctx, AV_LOG_ERROR, "FFT size %d not power of 2.\n", s->fft_size);
1756 return AVERROR_INVALIDDATA;
1759 ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R);
1760 ff_mpadsp_init(&s->mpadsp);
1762 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
1764 ff_thread_once(&init_static_once, qdm2_init_static_data);
1769 static av_cold int qdm2_decode_close(AVCodecContext *avctx)
1771 QDM2Context *s = avctx->priv_data;
1773 ff_rdft_end(&s->rdft_ctx);
1778 static int qdm2_decode(QDM2Context *q, const uint8_t *in, int16_t *out)
1781 const int frame_size = (q->frame_size * q->channels);
1783 if((unsigned)frame_size > FF_ARRAY_ELEMS(q->output_buffer)/2)
1786 /* select input buffer */
1787 q->compressed_data = in;
1788 q->compressed_size = q->checksum_size;
1790 /* copy old block, clear new block of output samples */
1791 memmove(q->output_buffer, &q->output_buffer[frame_size], frame_size * sizeof(float));
1792 memset(&q->output_buffer[frame_size], 0, frame_size * sizeof(float));
1794 /* decode block of QDM2 compressed data */
1795 if (q->sub_packet == 0) {
1796 q->has_errors = 0; // zero it for a new super block
1797 av_log(NULL,AV_LOG_DEBUG,"Superblock follows\n");
1798 qdm2_decode_super_block(q);
1801 /* parse subpackets */
1802 if (!q->has_errors) {
1803 if (q->sub_packet == 2)
1804 qdm2_decode_fft_packets(q);
1806 qdm2_fft_tone_synthesizer(q, q->sub_packet);
1809 /* sound synthesis stage 1 (FFT) */
1810 for (ch = 0; ch < q->channels; ch++) {
1811 qdm2_calculate_fft(q, ch, q->sub_packet);
1813 if (!q->has_errors && q->sub_packet_list_C[0].packet) {
1814 SAMPLES_NEEDED_2("has errors, and C list is not empty")
1819 /* sound synthesis stage 2 (MPEG audio like synthesis filter) */
1820 if (!q->has_errors && q->do_synth_filter)
1821 qdm2_synthesis_filter(q, q->sub_packet);
1823 q->sub_packet = (q->sub_packet + 1) % 16;
1825 /* clip and convert output float[] to 16-bit signed samples */
1826 for (i = 0; i < frame_size; i++) {
1827 int value = (int)q->output_buffer[i];
1829 if (value > SOFTCLIP_THRESHOLD)
1830 value = (value > HARDCLIP_THRESHOLD) ? 32767 : softclip_table[ value - SOFTCLIP_THRESHOLD];
1831 else if (value < -SOFTCLIP_THRESHOLD)
1832 value = (value < -HARDCLIP_THRESHOLD) ? -32767 : -softclip_table[-value - SOFTCLIP_THRESHOLD];
1840 static int qdm2_decode_frame(AVCodecContext *avctx, void *data,
1841 int *got_frame_ptr, AVPacket *avpkt)
1843 AVFrame *frame = data;
1844 const uint8_t *buf = avpkt->data;
1845 int buf_size = avpkt->size;
1846 QDM2Context *s = avctx->priv_data;
1852 if(buf_size < s->checksum_size)
1855 /* get output buffer */
1856 frame->nb_samples = 16 * s->frame_size;
1857 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1859 out = (int16_t *)frame->data[0];
1861 for (i = 0; i < 16; i++) {
1862 if ((ret = qdm2_decode(s, buf, out)) < 0)
1864 out += s->channels * s->frame_size;
1869 return s->checksum_size;
1872 const AVCodec ff_qdm2_decoder = {
1874 .long_name = NULL_IF_CONFIG_SMALL("QDesign Music Codec 2"),
1875 .type = AVMEDIA_TYPE_AUDIO,
1876 .id = AV_CODEC_ID_QDM2,
1877 .priv_data_size = sizeof(QDM2Context),
1878 .init = qdm2_decode_init,
1879 .close = qdm2_decode_close,
1880 .decode = qdm2_decode_frame,
1881 .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_CHANNEL_CONF,
1882 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,